The two bigger changes here are:
1- The http_client has been moved from the Session to the Browser, allowing
its connection pool to be re-used across multiple sessions
2- The browser now has a page_arena which is used for all page-level allocation
and which can be re-used between pages (currently retains 1MB of memory).
Previously, pages uses an arena that was tied to the lifetime of the page,
thus it could not be re-used.
Using the Bench allocator for zig-js-runtime, allocated bytes went from
1347037879 to 834932438 (in a RUNS=1000 of puppeteer demo).
Various other changes to try to simplify the API and remove the possibility
of invalid states. For example, session.newPage() now includes the logic for
page.start() so that there should now never be a page that wasn't started.
I don't know if FrameId is related to an <iframe>, and whether each Page has
1 implicit "frame". But, playwright seems to treat frameId and targetId as
interchangeable, and chrome seems to agree (at leas to some degree); chrome will
return a targetId and reuse that value for the frameId.
So the simplest solution is just to remove our concept of a frameId and use
targetId exclusively. This doesn't seem to cause any issues with puppeteer.
The getTargetInfo result must return a `targetInfo` key.
Here is an example returned by Chrome:
```json
{
"id": 16,
"result": {
"targetInfo": {
"targetId": "d93a1bbc-f906-4bbb-bb4d-a2285234b091",
"type": "browser",
"title": "",
"url": "",
"attached": true,
"canAccessOpener": false
}
}
}
```
Rely on inspector to send the result, otherwise we'll send 2 responses to the
same message (one ourselves and one from the inspector), which Playwright does
not like.
The TL;DR is that this commit enforces the use of correct IDs, introduces a
BrowserContext, and adds some CDP tests.
These are the ids we need to be aware of when talking about CDP:
- id
- browserContextId
- targetId
- sessionId
- loaderId
- frameId
The `id` is the only one that _should_ originate from the driver. It's attached
to most messages and it's how we maintain a request -> response flow: when
the server responds to a specific message, it echo's back the id from the
requested message. (As opposed to out-of-band events sent from the server which
won't have an `id`). When I say "id" from this point forward, I mean every id
except for this req->res id.
Every other id is created by the browser.
Prior to this commit, we didn't really check incoming ids from the driver. If
the driver said "attachToTarget" and included a targetId, we just assumed that
this was the current targetId. This was aided by the fact that we only used
hard-coded IDS. If _we_ only "create" a frameId of "FRAME-1", then it's tempting
to think the driver will only ever send a frameId of "FRAME-1".
The issue with this approach is that _if_ the browser and driver fall out of sync
and there's only ever 1 browserContextId, 1 sessionId and 1 frameId, it's not
impossible to imagine cases where we behave on the thing.
Imagine this flow:
- Driver asks for a new BrowserContext
- Browser says OK, your browserContextId is 1
- Driver, for whatever reason, says close browserContextId 2
- Browser says, OK, but it doesn't check the id and just closes the only
BrowserContext it knows about (which is 1)
By both re-using the same hard-coded ids, and not verifying that the ids sent
from the client correspond to the correct ids, any issues are going to be hard
to debug.
Currently LOADER_ID and FRAEM_ID are still hard-coded. Baby steps.
Previously, we could have multiple in-flight messages from the server to a
single client. This isn't safe and can lead to message interleaving. While
write / send are atomic, they are only atomic for the N bytes which they write,
which may not be the entire buffer. Consider this writeAll function:
```
pub fn writeAll(socket: socket_t, bytes: []const u8) !void {
var index: usize = 0;
while (index < bytes.len) {
index += try posix.write(socket, bytes[index..]);
}
}
```
If we're trying to send "abc123", this could take anywhere from 1 to 6 calls
to posix.write (it would take 6 calls, for example, if every call to
posix.write only wrote a single byte). Now if you're trying to write other data
to this same socket at the same time, messages _will_ get interleaved.
In order for this to work, the client now has a send_queue (doubly linked list).
When one message is sent, it sends the next.
In addition to the above change, the Client is now self-contained with respect
to its lifetime. This is necessary so that completions which come in AFTER our
concept of its lifetime ends, can still be processed. I think all types that
receive completions need to follow this model. This relies on the fact that
kqueue (which I know for a fact) and io_uring (which people seem to imply) handle
socket shutdown properly. It's still a bit messy because of timeout and not
wanting to wait until timeout to accept new connections, but needing to wait
until timeout to cleanup the client.
The self-contained nature of Client makes it difficult to test as a generic. I
removed Client(T). Tests now use real sockets. Some tests had to be removed
because they're too difficult to test over a real connection :(
